Elon Musk has threatened to nix the Twitter deal over too many bots on the platform. Here's why Musk is so fixated on fake accounts — and why they're so tricky to measure.

• Elon Musk has threatened to walk away from his Twitter buyout over bot accounts on the platform.
• Twitter says less than 5% of accounts are bots, but Musk and others argue it could be much higher.

For most of us, Twitter bots may seem like just another online pest. But for Elon Musk, they might be enough to scrap a multibillion-dollar deal.

The Tesla and SpaceX CEO has proposed buying and privatizing Twitter in what he views as a way to "unlock" its "extraordinary potential" as a global free-speech platform.

But since submitting a $44 billion buyout offer in April, Musk has become increasingly focused on spam and bot accounts on the platform. He said in May that the deal can't move forward unless there's proof that less than 5% of Twitter accounts are bots — on Tuesday, Musk accused Twitter of "actively resisting" his efforts to investigate fake accounts and threatened to pull out of the deal altogether.

Musk's fixation on bots has led observers to wonder if the issue is just an excuse to wriggle out of the deal following a Tesla stock slump. But bots do have real implications for ad revenue, and bot accounts are famously tricky to measure.

Here's what we know about Twitter's bot issues and what it could mean for the Twitter deal.

Why are there bots on Twitter and how long have they been around?

While Musk may be laser-focused on bots in 2022, they're nothing new for Twitter — in fact, bot accounts have existed on the platform since its founding in the mid-aughts.

Perhaps the most famous example of the power of Twitter bots was in 2016, when a Russia-linked troll farm unleashed over 50,000 bots in an effort to sway the outcome of the US presidential election.

But not every bot is created to run political interference. Rival companies may create bots that click on their competitors' ads in order to skew their rival's analytics. Or, bots might be designed to flood a site with traffic to cause a denial-of-service attack, Guy Tytunovich, the founder and CEO of cybersecurity firm Cheq, told Insider.

"Bots are created to mimic a human being's behavior in order to benefit the proprietor of said bot," Tytunovich said.

How many bots are there on Twitter?

Measuring the exact number of Twitter bots is a thorny issue, and everyone seems to have a different estimate.

Twitter has long maintained that fewer than 5% of the accounts on its platform are bots. That figure is the result of quarterly human reviews of "thousands of accounts, that are sampled at random, consistently over time," Twitter CEO Parag Agrawal tweeted in May.

But Musk is arguing that Twitter's figure is low — very low. He said during a recent episode of the "All-In" podcast that he thinks the number is at least 20%. "Currently what I'm being told is that there's just no way to know the number of bots," said Musk at the time. "It's like, as unknowable as the human soul."

Some outside firms have taken a stab at figuring it out. Data analytics firm GlobalData found that 10% of Twitter's active accounts are posting spam, which was based on a sample of 22,000 tweets.

At Cheq, Tytunovich's team analyzed 5.21 million visits to its customers' websites that originated from Twitter and found that 11.71% of those visits were by bots. Cheq found that the majority of those accounts were malicious bots and scrapers, which collect large swaths of data from web pages.

But Cheq's analysis was based solely on the bots that actually clicked on an ad and were sent to another website, which means that bots that didn't click on ads weren't counted.

"I'm imagining that the number is far higher in terms of the bots that are on Twitter," Tytunovich said.

Why is it so hard to measure?

Part of the challenge of measuring bot accounts is that the concept of spam itself is subjective, Sidharth Kumar, GlobalData's senior data scientist, wrote in a blog post about the firm's findings.

"It is almost impossible to confirm the identity of the entity behind a tweet handle," he wrote. "Additionally, the definition of spam account may differ for everyone. Incessant tweeting of non-original content can be considered spam, but some may choose to see it as a very active user sharing articles/opinions."

Agrawal has made a similar point: "FirstnameBunchOfNumbers with no profile pic and odd tweets might seem like a bot or spam to you, but behind the scenes we often see multiple indicators that it's a real person," he tweeted last month.

Twitter says it's impossible for an outsider to determine whether an account is, in fact, a bot, because doing so requires user data like IP addresses, phone numbers, geolocations, and how the account behaves when it's active.

Why does Musk care about bots so much?

The short answer: advertising revenue.

Musk's team is arguing that if there are more bots on Twitter than the company is saying there are, the proportion of real users who will actually be shown ads is smaller, which would impact ad revenue.

"How do advertisers know what they're getting for their money?" Musk tweeted in May in response to Agrawal. "This is fundamental to the financial health of Twitter."

But some speculate that spam may just be a loophole for Musk to get out of the deal. Although he waived his right to comb through Twitter's finances before purchasing the company, his purchase agreement is void if he's able to prove that Twitter misled him.

Which means that if Musk can prove that more than 5% of Twitter's users are bots, he may be able to walk away altogether.

What happens next?

Twitter has so far refused to hand its internal user data over to Musk, but that may be about to change: Twitter has reportedly given in to Musk's demands and will allow him access to its "firehose" of internal data, the Washington Post reported Wednesday.

While it may seem like a victory for Musk, Twitter's leaders are reportedly skeptical that this data will prove useful to his investigation — due to a wrinkle in how bots are measured at Twitter. When Twitter says less than 5% of accounts are bots, it's measuring a random sample of mDAU, or monetizable daily active users — as in, users who were logged in on a given day through Twitter's website, or a Twitter application that's able to show ads.

The New York Times reported that this means in order to investigate Twitter's claims, Musk will most likely require a team of experts and massive computing capacity to process that "firehose" of data — which includes more than 500 million tweets sent every day.